What the best AI models can do, what they cost, how open and closed models compare, and where capability is heading. Refreshed every morning at 07:00 AEST.
Loading data…
Today's read
Leaderboard: open vs closed
Intelligence score: Artificial Analysis Intelligence Index, a composite of 10 hard evaluations. ECI: Epoch Capabilities Index. Cost per task: what it cost Artificial Analysis to run one index task. Prices are US$ per million tokens.
Capability vs cost
Intelligence score against US$ per index task (log scale) · up and left is better value
ClosedOpen weights
Headline benchmarks
† vendor-reported; others independently measured
Where capability is heading
Closed frontierOpen frontierProjection range
Closed frontier projected at the low–high growth rate. Open frontier projected as the closed frontier shifted by the open-model lag. Both are trend estimates, re-fitted daily as new models land.
Milestone dates
central estimate, with range
How long a task AI can finish on its own
METR 50% time horizon (task length for a skilled person) · dots measured · band projected
Subscriptions vs API: what heavy use really costs
plan prices are published; API-equivalent value and agent-hour capacity are estimates (vendors don't publish token budgets)